A Relaxation Argument for Optimization in Neural Networks and Non-Convex Compressed Sensing

G. Welper

It has been observed in practical applications and in theoretical analysis that over-parametrization helps to find good minima in neural network training. Similarly, in this articl…